For each observation, compares the mean distance to the rest of its own
cluster (a) with the smallest mean distance to any other cluster (b),
giving a width of (b - a) / max(a, b). Widths near 1 indicate a
well-placed observation, near 0 one on a boundary, and negative ones an
observation that sits closer to a different cluster.
Arguments
- d
A stats::dist object, or a numeric matrix or data frame, in which case Euclidean distances are computed with
shoal_dist()first.- cluster
A clustering result from this package, or an integer vector of cluster IDs.
NAentries, the noise points of the density-based algorithms, are excluded along with their distances.
Value
A data frame with one row per observation and columns cluster,
neighbour (the nearest other cluster) and width. The mean width is
attached as the avg_width attribute.
Details
An observation alone in its cluster is given a width of 0 by convention, since it has no within-cluster distances to average.
See also
shoal_metrics() for indices computed from the data instead.
Examples
x <- as.matrix(iris[, 1:4])
fit <- shoal_kmeans(x, k = 3L)
sil <- shoal_silhouette(shoal_dist(x), fit)
attr(sil, "avg_width")
#> [1] 0.552819